REvIEwS

Synthetic lethality as an engine for cancer drug target discovery

Alan Huang1, Levi A. Garraway2,4, Alan Ashworth3 and Barbara Weber 1* Abstract | The first wave of genetically targeted therapies for cancer focused on drugging gene products that are recurrently mutated in specific cancer types. However, mutational analysis of tumours has largely been exhausted as a strategy for the identification of new cancer targets that are druggable with conventional approaches. Furthermore, some known genetic drivers of cancer have not been directly targeted yet owing to their molecular structure (undruggable oncogenes) or because they result in functional loss (tumour suppressor genes). Functional genomic screening based on the genetic concept of synthetic lethality provides an avenue to discover drug targets in all these areas. Although synthetic lethality is not a new idea, recent advances, including CRISPR-​based gene editing, have made possible systematic screens for synthetic lethal drug targets in human cancers. Such approaches have broad potential to drive the discovery of the next wave of genetic cancer targets and ultimately the introduction of effective medicines that are still needed for most cancers.

Cancer is a disease of the genome, and almost all can­ cancer24,25, EGFR mutations and ALK translocations in cers have multiple genetic lesions that must be addressed non-​small-cell lung cancer (NSCLC)26–35 and numerous to develop curative combination therapies1. A reduc­ others36. However, our ability to make further progress tionist view of the hallmarks of cancer suggests that tar­ with genetically targeted cancer therapy has been limited geting oncogenic drivers, tumour suppressor gene loss by two main issues. First, although partial responses to and the underlying mechanisms by which cancer cells targeted therapies in selected patient populations are evade immune destruction are the minimum that will common, converting those partial responses to dura­ be required for cures1. ble complete responses, which is needed for cures, will Completion of the Human Genome Project in 2003 require combination regimens that have been challeng­ (refs2,3) followed by advances in sequencing technol­ ing to define. Additionally, although DNA sequencing ogy and the analysis of thousands of human tumours4–7 technology has enabled the identification of most, if enabled discovery of the first generation of genetically not all, oncogenes that arise as a consequence of genetic targeted cancer therapies, which, even as single agents, alterations37–39, they represent a relatively small percent­ changed the lives of many people with cancer. Imatinib, age of genes that are relevant in cancer, not all of which which targets the BCR–ABL fusion tyrosine kinase, has are druggable with conventional approaches. Proteolysis-​ extended the median survival of patients with chronic targeting chimaeras (PROTACS) and other protein myelogenous leukaemia to more than 10 years8–11. degradation approaches are likely to change the definition Imatinib also inhibits KIT and is effective in treating of ‘druggable’ in the coming years40,41, but independent of 1Tango Therapeutics, KIT-​mutant gastrointestinal stromal tumours, with advances that may redefine druggability, identification Cambridge, MA, USA. response rates of up to 50% and a median progression-​ of the next wave of cancer drug targets requires more than 2Eli Lilly and Company, free survival of approximately 18 months12–15. The deep sequencing of multiple tumours. CRISPR-​enabled Indianapolis, IN, USA. BRAF inhibitor vemurafenib16, followed a few years functional genomic screening platforms are powerful tools 3UCSF Helen Diller Family later by dabrafenib17 and encorafenib18, transformed for this application. Comprehensive Cancer BRAF-​mutant melanoma from an untreatable, rapidly As enthusiasm for targeted cancer therapy waned Center, San Francisco, CA, USA. progressive malignancy to a disease in which more than somewhat in the face of its limitations, progress harness­ 50% of patients have meaningful clinical responses19,20 ing the immune system to treat cancer allowed another 4Present address: Roche/Genentech, South and when combined with a MEK inhibitor, have a wave of clinically meaningful responses, and, in some San Francisco, CA, USA. median progression-​free survival of approximately cases, cures. Many patients with melanoma, regardless 21–23 42–44 45 *e-​mail: [email protected] 12 months with limited toxicity . There are multiple of BRAF mutation status , renal cell carcinoma , 46–49 50 https://doi.org/10.1038/ other examples of clinical successes, including drugs NSCLC and a number of other cancers have clin­ s41573-019-0046-z that target amplified ERBB2 (encoding HER2) in breast ically meaningful responses to checkpoint inhibitors

Nature Reviews | Drug Discovery Reviews

such as anti-​programmed cell death 1 (PD1) or anti-​ decade. Here we analyse the evidence that many cancer PD1 ligand 1 (PD-​L1), as well as in some cases antibod­ targets remain to be discovered using a CRISPR-​based ies against cytotoxic T lymphocyte antigen 4 (CTLA4), functional genomic screening and how this approach alone or in combination with another checkpoint can lead to the discovery of additional targets based on inhibitor51,52. These important successes further damp­ loss of tumour suppressor genes. We also highlight the ened enthusiasm for genetically targeted therapies and importance of genetic context in designing target discov­ spurred a wave of clinical investigation in a broad range ery strategies, and analyse the technical considerations of tumour types with many putative immune cell tar­ for scalable synthetic lethal target discovery, including gets. This next wave of immunotherapy agents has yet to the relative benefits of various CRISPR-​based tools prove as effective as anti-​PD1 and anti-​PD-L1 agents53, and libraries, discussing inhibition of protein arginine and it remains to be determined whether immunother­ N-​methyltransferase 5 (PRMT5) in cancers with dele­ apies will do what the initial targeted therapies failed to tion of S-​methyl-5′-thioadenosine phosphorylase do: to produce durable complete responses and cures in (MTAP) as an example of targeting synthetic lethality large numbers of patients with cancer. beyond poly(ADP-​ribose) polymerase (PARP) inhibitors Therefore, although advances in both targeted therapy in BRCA1-mutant and BRCA2-mutant contexts. Finally, and immunotherapy have improved cancer treatment we discuss the applications of CRISPR-​based screening for many people in remarkable ways, barriers to target­ with targeted drugs for novel combination discovery and ing tumour suppressor gene loss, identifying context-​ in vivo screening for non-​cell-autonomous mechanisms. dependent driver genes that are not marked by genetic alterations (which result in non-​oncogene addiction54,55, Synthetic lethality called here ‘unmarked oncogenes’), designing the next The genetic concept of synthetic lethality was first generation of novel combinations and exploring the described in Drosophila in 1922 (refs56,60–62). Fruit flies largely unknown genetics of tumour-​intrinsic immune with individual abnormal eye phenotypes (phenotypes evasion have stalled progress. However, the genetic prin­ that were attributed to either Bar or glass mutations) ciple of synthetic lethality coupled with the power of could survive and reproduce but viable offspring with a CRISPR-​based functional genomic screening technology combination phenotype were never observed56. We now offers a path forward. know that Bar and glass encode transcription factors Synthetic lethality, initially described in Drosophila expressed in the eye, an extension of the central nervous as recessive lethality56, is classically defined as the setting system, and are involved in embryonal development. We in which inactivation of either of two genes individually can therefore postulate that loss of both genes simultane­ has little effect on cell viability but loss of function of ously results in neural development defects not compat­ both genes simultaneously leads to cell death. In can­ ible with life, although the specific lethality mechanism cer, the concept of synthetic lethality has been extended of this synthetic lethal pair has not been studied directly. to pairs of genes, in which inactivation of one by dele­ The concept was subsequently also shown to be relevant tion or mutation and pharmacological inhibition of the in yeast63, and eventually was proposed as a basis for other leads to death of cancer cells whereas normal cells drug discovery for human disease by Hartwell, Friend (which lack the fixed genetic alteration) are spared the and colleagues 20 years ago57. Hartwell, and subsequently effect of the drug. In the most straightforward applica­ Kaelin, proposed that novel targets for cancer therapeu­ tion, this means identifying targeted therapies that kill tics could be discovered by exploiting the concept of syn­ cancer cells that lack a specific tumour suppressor gene thetic lethality, in which one member of a synthetic lethal but spare normal cells. In addition to this (conceptu­ pair is a gene product with a cancer-​specific mutation ally) simple application, the tools needed to discover and the second gene product is the drug target57,58. synthetic lethal interactions in human cancer cells can The systematic identification of synthetic lethal pairs be applied in multiple ways to identify a number of other relevant to human disease was initially limited to loss-​ types of cancer drug targets. of-function screens in model organisms. For example, In this Review, we discuss how the genetic con­ genetic interaction maps in yeast were readily gener­ cept of synthetic lethality paired with CRISPR-​based ated by genetic screening and crossing of knockout functional genomic screening can be applied to iden­ strains57,63–65, but the utility of these findings for cancer tify the next generation of effective cancer drugs and drug discovery relies on the existence of relevant homo­ combinations. Although much has been written about logues in human cells. The advent of RNAi technology synthetic lethality in cancer since Hartwell, Friend and allowed the broader application of these concepts to colleagues raised the idea in 1997 (refs57–59), application human cell line systems, and a number of subsequent of the concept to cancer drug discovery has been largely screening efforts in cancer cells were undertaken66. aspirational. This is largely because of the limitations Proof-​of-concept classical synthetic lethal screening of using yeast and Drosophila as model organisms for using small interfering RNA (siRNA) to identify depen­ human disease and the limitations of previous genera­ dencies conferred by loss of the tumour suppressor gene tions of genetic tools, such as RNA interference (RNAi), VHL (encoding von Hippel–Lindau disease tumour Genetic context for mammalian studies. The widespread availability of suppressor) in clear cell renal carcinoma was described Histology and genetic CRISPR-​based tools and the increasingly varied ways in 2008 (ref.67). Further application of systematic RNAi architecture that define a specific set of cancer patients they can be used has created an inflexion point that has screening led to the discovery of additional synthetic (for example, patients with the potential to dramatically alter cancer target discov­ lethal pairs, including members of the SWI/SNF chro­ BRCA1-mutant ovarian cancer). ery and create a wave of new therapeutics in the next matin remodelling complexes such as SMARCA2 and

www.nature.com/nrd Reviews

ab100 76 Hazard ratio, 0.27 (95% Cl, 0.17–0.41) rucaparib, niraparib and talazoparib) . Of note, PARP BRCA-mut Normal P<0.001 inhibitors seem more effective in BRCA1-mutant and cancer cell cell 75 BRCA2-mutant ovarian cancers than in breast cancers with these mutations, which raises the consideration of Niraparib additional ‘genetic context’, a critical factor in design­ ee survival (%) 50 ing functional genomic target discovery screens, as discussed later.

+ PARPi + PARPi ession-fr 25 Placebo

ogr Many cancer targets to be discovered Pr Large-​scale gene-​knockout studies across many genetic 0 0 2 4 6 8 10 12 14 16 18 20 22 24 contexts are now being used to map synthetic lethal Cell death No effect Months since randomization interactions in human cancer cells — first using short hairpin RNA (shRNA)-based approaches and more Number at risk recently using CRISPR technology, which has elim­ Niraparib 138 125 107 98 89 79 63 44 28 26 16 31 inated many of the technical hurdles of RNAi-​based Placebo 65 52 34 21 12 86222110 functional genomic screens83–85 (Box 1). Project DRIVE Fig. 1 | synthetic lethality: a genetic concept reduced to clinical practice. Schematic (conducted at Novartis) and Project Achilles (con­ (part a) and phase III clinical data197 (part b) showing a clinical example of the utility of ducted at the Broad Institute) use the Cancer Cell Line synthetic lethality for drug development. Poly(ADP-ribose)​ polymerase inhibition (PARPi), Encyclopedia — a large panel of human cancer cell lines which has minimal effect in normal cells with wild-type​ BRCA gene function, causes that represent multiple cancer types86 — to create a tumour cell-specific​ death in BRCA gene-​mutated tumour cells, significantly extending catalogue of essential genes and synthetic lethal inter­ the progression-free​ survival of patients with germline mutations in BRCA genes. actions87–91. The Sanger Institute has taken a similar 187 BRCA-mut,​ BRCA gene mutant. Part b from ref. , N. Engl. J. Med., Mirza et al., Niraparib approach in 324 human cancer cell lines from 30 cancer maintenance therapy in platinum-sensitive,​ recurrent ovarian cancer, 375, 2154–2164. types (Project Score)92. The number of novel druggable Copyright © 2016, Massachusetts Medical Society. Reprinted with permission from Massachusetts Medical Society. targets that have been nominated and fully validated by Project Achilles and Project DRIVE has thus far been limited, with PRMT5 (refs72–74) and Werner syndrome SMARCA4 (refs68–70) and ARID1A and ARID1B (ref.71), ATP-​dependent helicase (WRN)92–95 being the best in addition to the PRMT5–MTAP interaction72–74 examples. However, Project Score focused specifically discussed in detail later in this Review. on identification of additional novel drug targets, and The recent success of PARP inhibitors in BRCA-​ has provided convincing evidence that many undiscov­ mutant ovarian cancers is the first clinical example of ered targets indeed exist, that most are context depen­ using synthetic lethality to target tumour suppressor dent, and that they can be discovered with a functional gene loss75–79 (Fig. 1). The basis for this finding is that both genomics approach92. PARP and BRCA1 and BRCA2 are components of effi­ The Sanger Institute effort, recently described by cient DNA repair. This interaction makes tumour cells Garrett and colleagues, includes 941 CRISPR–Cas9 with mutations in BRCA1 or BRCA2 sensitive to PARP screens in 339 cell lines from the Cell Model Passport inhibition, driving efficacy. Normal cells, which have at collection using a genome-​scale library targeting least one copy of BRCA1 or BRCA2, are largely spared, ~18,000 genes92. The final analysis included 324 of those which limits toxicity. It now has been well described cell lines. Genes required for cell fitness (called ‘core fit­ that all PARP inhibitors that have reached clinical stages ness genes’ or ‘essential genes’) across the majority of cell have both catalytic inhibitory and DNA trapping activ­ lines were deemed unlikely to be good drug targets due ity80,81, which has led to some controversy regarding the to a narrow therapeutic index. In total, 7,470 targeted mechanism of lethality. Whereas the DNA trapping genes (41%) altered cell viability in at least one cell line activity of PARP inhibitors clearly enhances the effect of (Box 2). Good drug targets, likely to be represented by genetic knockdown of PARP, the activity of PARP inhib­ fitness genes restricted to specific molecular contexts itors in tumours with BRCA1 or BRCA2 mutations is or histologies, were defined as a subset of those genes tightly linked to loss of function of the gene products. with fitness effects in 12 or fewer of the 13 cancer types Furthermore, it has been shown that PARP inhibitors included in the screen. With a defined median of 1,459 with markedly different DNA trapping potencies have fitness genes per cell line, a median of 553 genes were comparable activity as measured by growth inhibition at considered pan-​cancer fitness genes and 866 were nom­ maximum tolerated doses in xenograft models of BRCA1- inated as cancer type-​specific genes. To further refine mutant triple-​negative breast cancer82. As PARP inhi­b­ priority drug targets, genes were scored for various itors are used at doses that maximize both PARP cata­ measures of the effect of gene ablation, including extent lytic inhibition and DNA trapping activity, the relative of the effect, levels of expression of the target gene and importance of those functions in clinical response mutational status (70% of the priority score), as well as remains unknown. Therefore, although the exact mecha­ evidence of a genetic biomarker associated with the tar­ nism underlying PARP–BRCA1 and PARP–BRCA2 get dependency and the somatic alteration frequency of synthetic lethality remains unclear, DNA damage the target gene in human cancers (the remaining 30% repair as the basis for the interaction remains undis­ of the priority score). With this paradigm, 617 priority puted. The FDA has approved four PARP inhibitors for cancer type-​specific targets and 92 priority pan-​cancer use in patients with BRCA-​mutant cancers (olaparib, targets were defined. Most of the cancer type-​specific

Nature Reviews | Drug Discovery Reviews

priority targets were identified in two or fewer cancer United States96 and, whereas there are more than 90 types (n = 457; 74%), emphasizing the importance of NSCLC cell lines, there is only one cell line each rep­ context. Finally, a tractability filter for the development resenting ALK translocation or MET amplification, of small molecules and therapeutic antibodies was and there are no cell lines with MET exon 14 skipping applied. Tractability group 1, which included 40 genes mutations87,90. ALK translocations are present in less (6%), included all targets of drugs approved or already than 5% of lung cancers overall97, but they are dispro­ in development. A large number of priority targets, portionally represented (40%) in younger non-​smokers 277 genes (44%), fell into tractability group 2, defined (<40 years) with lung adenocarcinoma98. As in the case of as targets with no known development efforts but with ALK mutations, genetic alterations in MET are similarly evidence supporting target tractability. Finally, the lar­ uncommon in lung cancer overall but represent a very gest number of putative targets, 311 (48%), were deemed important therapeutic target in a similar proportion of untractable by conventional means. Although these patients97. Second, the intentional heterogeneity of the data do not suggest that all genes with a fitness effect are Cancer Cell Line Encyclopedia collection introduces a good drug targets, or that all undiscovered targets mathematical challenge to identifying even strong signals are represented in this analysis, they provide strong evi­ when they are present in a small subset of the collection. dence for the presence of a large target space remaining This is the same problem encountered in clinical trials to be explored and many paths to discover effective new of targeted agents conducted without patient selection: if treatments for patients with cancer. the number of patients in the trial with the genetic con­ Given the evidence from the Sanger analysis that a text necessary for response is small, even a large signal large number of druggable cancer targets remain to be in individual patients will be diluted by non-​responders, discovered, why have so few emerged from other large-​ and the trial will fail. An analogous problem occurs when scale experiments thus far? The answer likely lies in one is analysing large heterogeneous panels of cell lines: both the primary aim of Project Achilles and similar a strong signal from a small number of cells will be lost. projects87,92 and their design. Although novel drug tar­ This problem was addressed analytically in Project Score, get discovery is a potential benefit of Project Achilles, revealing the strength of the effect: of the 628 priority the overarching goal is to increase our understanding of targets identified in the analysis, 56% had a fitness effect cancer biology. The data obtained are extremely valuable in only one cancer type and an additional 18% had a for many applications, but efficient and comprehensive fitness effect in only two cancer types. Third, the readout drug target and combination discovery requires some for Project DRIVE and Project Achilles screens is guide refinements tailored to specific applications. RNA (gRNA) ratios, which reflect growth rate changes First, although Project Achilles and Project DRIVE and/or cell death from in vitro monocultures, eliminat­ have included more than 300 cancer cell lines so far, ing the ability to discover non-​cell-autonomous mecha­ there are many cancer subtypes insufficiently repre­ nisms such as immune evasion. Finally, the screens are sented for context-​specific analyses and some histologies conducted in vitro with CRISPR or shRNA libraries as are not represented at all. For example, in the collection the sole perturbation, and thus the data cannot be used there are no cell lines for human papillomavirus-​positive to identify novel combination therapies because only one head and neck cancer, which represents more than half gene is perturbed in each cell and no drugs are used in of all newly diagnosed head and neck cancer cases in the combination with the genetic perturbation. These last two issues cannot be addressed analytically in any of the Box 1 | Big data approaches to synthetic lethal drug target discovery large datasets now available and require experimental designs specific to those types of targets. Project Achilles is a functional genomic screening initiative from the Broad Institute with the goal of creating a genome-wide​ catalogue of tumour vulnerabilities associated Following the successful application of loss-of- with genetic and epigenetic alterations. Short hairpin RNA (shRNA) screening and now function screening across large cell line panels and the CRISPR-​based screening have been applied to cancer cell lines at the genome scale discovery of CRISPR, a number of biotechnology com­ to interrogate gene essentiality, providing the foundation for a cancer dependency panies have been founded in recent years to identify and map88–91. Project DRIVE (deep RNA interference (RNAi) interrogation of viability effects prosecute novel synthetic lethal drug targets, and it is in cancer) is a large-scale​ RNAi screen of almost 400 cancer cell lines led by Novartis to likely that a number of pharmaceutical companies have define cancer dependency genes. Project DRIVE was designed to overcome the inability started to use this approach for target discovery when of RNAi screens to distinguish between on-target​ and off-target​ effects when using low relevant model systems are available. The first wave of numbers of shRNAs per gene and to increase the statistical power to describe molecular targeted therapies from these efforts is emerging, and it correlates of knockdown effects. In a large-scale​ robotics approach, a lentiviral library seems likely that the initial clinical trials from these efforts targeting 7,837 human genes was produced with a median of 20 shRNAs per gene (compared with the standard use of three to five shRNAs per gene) and was used to will start to enrol patients within the next few years. screen 398 cancer cell lines in a pooled format87. Project Score is an initiative from the Sanger Institute to profile the genetic The importance of genetic context dependencies of 324 cancer cell lines across 19 tissues and 30 cancer types using a As noted already, considering genetic context is impor­ whole-​genome CRISPR library targeting approximately 18,000 genes92. These data tant to avoid signal dilution due to heterogeneity in both were specifically analysed to determine the number of novel, tractable drug targets clinical trial design and experimental design. The two that remain to be discovered, convincingly demonstrating that the number is likely in are tightly linked: the genetic context selected for target the hundreds. This dataset can be used to identify potential drug targets that are discovery should form the basis for a patient selection active in a predictable subset of cancer cell lines, as exemplified by discussion in the strategy in clinical development. In addition, many syn­ publication of Werner syndrome ATP-dependent​ helicase (WRN) as a target in thetic lethal pairs are likely to be context dependent, with microsatellite instable tumours. alterations in other genes in specific contexts altering

www.nature.com/nrd Reviews

99,101 Box 2 | crisPr screen in a large cancer cell line panel suggests many more reports suggest that EGLN inhibitors — currently drug targets to be discovered used clinically to treat anaemia — may be effective in treating women with ovarian clear cell carcinoma. The 350 350 A panel of 324 cell lines from therapeutic effect may be most effective in an ARID1A-​ the Sanger Project Score was mutant subset, but that hypothesis has not been tested screened using a genome-​ 300 300 in clinical trials. scale CRISPR library and then analysed to nominate those Another recent example of a synthetic lethal drug tar­ genes important for cellular get discoverable only through analysis of cell lines with 250 250 fitness. Of the genes tested, a specific genetic context is DNA polymerase θ (POLQ) 41% impacted growth of at in BRCA1-mutant and BRCA2-mutant cancers. POLQ least one cell line (7,470 fitness 200 200 is a low-​fidelity DNA polymerase that participates in genes represented in the alternative non-​homologous end joining104–107, a critical figure), with most of these 50% cell lines pathway for the repair of DNA double-​strand breaks in genes impacting 150 150 tumours with defective homologous recombination108,109. a minority of the cell lines POLQ knockdown reduces cellular survival both on (falling below the 50% of cell its own and in combination with PARP inhibitors in lines mark), suggesting context 100 100 specificity of the Number of dependent cell lines a BRCA1-mutant and BRCA2-mutant context but not a 108,109 dependency92. Strong drug wild-​type context . This target was not identified by 87,89 targets would be those that 50 50 the original Achilles and Project DRIVE analyses , but selectively kill a subset of cell is mentioned in the Sanger analysis92. This is probably lines with a predictable due to the fact that there was an under-​representation of genetic feature 0 0 BRCA1-mutant and BRCA2-mutant cell lines, but then 1 10 102 103 0 20 40 60 80 and harbour a conventionally this target became easily discoverable in a small panel Number of genes Genes (%) druggable protein domain. of curated BRCA1-mutant cancer cell lines paired with Figure reproduced with permission from ref.92, Springer Nature Limited. wild-​type BRCA1 isogenic derivatives110. KRAS mutations provide another example of the the functional interaction of the synthetic lethal pair. We complexity of context: cell lines and tumour models with define ‘genetic context’ as the driver mutation, tissue of G12C or G12A mutations in KRAS are very sensitive origin (histology) and other functional genetic lesions to tyrosine-​protein phosphatase non-​receptor type 11 that make up subgroups of cancers. The importance of (PTPN11) inhibitors, whereas those with mutations at genetic context is exemplified by the discovery of the G13 and Q61 are not111,112. This occurs because onco­ selective dependency of ovarian clear cell carcinoma genic G12 variants, but not G13 and Q61 variants, are and melanoma cell lines on the oxygen sensor EGLN1 dependent on PTPN11-mediated GTP loading to pro­ (ref.99). EGLN1 is a member of the EgIN family of prolyl mote downstream signalling111. Given that G12C and hydroxylases that regulates levels of hypoxia-​inducible G12A mutations are enriched in NSCLC but are rare in factor 1α (HIF1α) via hydroxylation and subsequent colorectal and pancreatic cancer, there may be differ­ degradation by VHL100. Hahn and colleagues linked a ences in sensitivity to PTPN11 inhibition on the basis cancer dependency to HIF1α upregulation, consistent of histology. Moreover, DRIVE data analysis suggests with the role of EGLN1 in regulating cellular HIF1α that cells with KRAS mutations that co-​occur with levels. They postulated that pharmacologic inhibition SMARCA4 or KEAP1 loss of function are less depen­ of EGLN1, which stabilizes HIF1α, reduces cellular fit­ dent on KRAS than those with wild-​type SMARCA4 ness, leading reduced proliferation and cell death99. In a or KEAP1 function87. These data again highlight the more focused analysis of just the ovarian carcinoma cell importance of considering additional genetic context lines that were included in Project Achilles, Briggs et al. in target discovery, and further suggest that in some identified ARID1A as an additional dependency101. cases mutation subtyping may be needed for patient In this analysis, it is not possible to separate the effects selection. of HIF1α upregulation and ARID1A mutation as all cell lines with ENGL1 dependency are characterized by Large-scale​ target discovery approaches HIF1α upregulation, and most of those have ARID1A As noted already, functional genomic screens, specif­ mutations. ARID1A is mutationally inactivated in ~60% ically high-​throughput loss-​of-function screens that of ovarian clear cell carcinomas102,103 and is frequently identify the pairwise effects of synthetic lethal gene mutated in other histologies but, of note, analysis of the pairs, have been envisioned as a path to identifying novel Project Achilles dataset in its entirety does not identify targets and combinations for many years58,59; however, any druggable ARID1A synthetic lethal interactions pre-​CRISPR technology was not sufficiently robust for (ARID1B, a paralogue of ARID1A, is a strong synthetic these applications. Before the discovery of CRISPR113–116, lethal partner with ARID1A71 but is not considered con­ RNAi was the standard tool for loss-​of-function genetic ventionally druggable owing to lack of enzymatic activ­ screens in mammalian cells117,118. Although initially ity or previously targeted domains). Briggs et al. showed very promising, it rapidly became clear that RNAi lacks that pharmacologic inhibition of EGLN1 selectively kills the specificity for high-​throughput applications. This ARID1A-​mutant ovarian cancer cells; thus, it is plausi­ problem is inherent to the technology, as stretches of ble that the combination of these two abnormalities is shRNA and siRNA sequences (called ‘seed sequences’) responsible for the dependency101. The data from both can bind to and downregulate mRNAs unrelated to the

Nature Reviews | Drug Discovery Reviews

Isogenic cell line pairs gene of interest, resulting in multiple, largely unavoid­ now sometimes called ‘CRISPR cutting’. These remarkable 119–121 Cultured cell lines genetically able, false positive hits in genetic screens . These advances opened the door to a variety of modifications, engineered to have only a off-​target effects can now be predicted and computa­ as well as the discovery of additional editing enzymes single genetic difference tionally filtered to improve interpretation of existing such as the endonuclease Cpf1 (refs128,129). Many of these between them. shRNA screening data89; however, unless 15–20 shRNAs emerging genomic engineering approaches have a role per gene are used for screening, identifying true hits in cancer target discovery, as described in more detail in from shRNA screens remains challenging and largely Box 3. As additional genetic tools and modifications impractical without large-​format robotics87. By con­ emerge, target discovery approaches will continue to be trast, CRISPR is a highly specific, efficient and scalable enhanced. genome-​editing technology markedly outperforming RNAi-​based reagents83–85,122 and can be applied in high-​ Strategies for CRISPR library selection. When one throughput screens to discover novel drug targets123,124 is considering functional genomic target discovery using multiple related approaches. screens using CRISPR-​based tools, the primary deter­ The initial discovery of CRISPR in microorgan­ minants of size, cost and time are the number of cell isms was based on the native bacterial enzyme Cas9 lines to be screened and the library size. The determi­ which excises segments of foreign DNA from the host nation of the number of cell lines largely depends on genome125–127. The utility of this powerful system in the purpose of the screen: useful information can be human cells was quickly realized with the development obtained with just one cell line or one isogenic pair, of gene-​editing tools that pair stable expression of bac­ particularly when CRISPR screening is combined with terial Cas9 with sequence-​specific gRNAs that guide a pharmacologic inhibitor, or hundreds of cell lines, the enzyme to excise precise DNA fragments from the as with Project DRIVE, Project Achilles and Project human genome and at genome scale if desired114–116, Score87,90,92,130. Isogenic cell line pairs have the advantage

Box 3 | crisPr technologies that enable synthetic lethal drug target discovery crisPr CRISPR technology uses a guide RNA with a target sequence of 20 base pairs in length to direct Cas9 to sequence-​ specific regions of the genome, resulting in the DNA cuts that lead to gene product loss of function113–115. In human cells, such an approach can be leveraged to systematically study the functional effect of loss of each gene in the genome123,124. In addition to Cas9, Cpf1 is a class 2 CRISPR system that relies on a single RNA-guided​ nuclease effector128,129. crisPr interference CRISPR interference (CRISPRi) uses catalytically inactive Cas9 (‘dead Cas9’, dCas9) fused to a Krüppel-associated​ box (KRAB) protein domain that interferes with transcription, suppressing gene expression rather than inducing double-​ strand DNA breaks172–174. CRISPRi has the advantage of more closely mimicking the effect of a pharmacologic inhibitor, which incompletely suppresses activity. As a result, CRISPRi may allow differentiation between the enzymatic effect of a gene product versus a potential scaffolding effect, the latter not being amenable to pharmacologic inhibition. However, the number of poorly characterized transcription start sites in the genome limits this application and has the potential to introduce false negative results in a large genomic screen. combinatorial crisPr CRISPR systems that allow interrogation of more than one gene per cell will be particularly valuable in defining novel drug combination regimens. The Cas9 and Cpf1 systems may both be used to this end. The Cpf1 system allows one-step​ direct cloning of concatenated gDNA and seems to have a substantially lower rate of recombination in excising the guide RNAs than Cas9 owing to the vector design175–180. Combinatorial CRISPR also can be applied for high-throughput​ isogenic testing, with the first position of the vector targeting a tumour suppressor gene and the second position targeting the druggable genome. This approach has the potential to dramatically scale the identification of synthetic lethal targets using well-controlled​ isogenic systems. Base editing Base editing also uses dCas9; however, the transcriptional repressor is replaced by a DNA deaminase, which results in base pair modifications that are precisely targeted181–184. This technology is particularly useful for rapid generation of isogenic pairs of cell lines, for example, changing the endogenous KRAS allele from mutant to wild type, or vice versa. Such a system is well controlled in comparison with traditional methods of overexpressing an exogenous expression construct and could therefore be very powerful for synthetic lethal-based​ target discovery. rNA targeting C2c2, Cas13b and Cas13d are the latest CRISPR systems reported185–191. These enzymes recognize and cut mRNA, making possible both reversible and scalable target gene modulation. Such an approach mimics small-molecule​ or antibody drug modes of action better than DNA editing, as pharmacologic inhibitors rarely fully inhibit the target gene product, suggesting these will be powerful tools for drug target discovery. single-​cell sequencing of pooled crisPr screens The complexity of the readout of pooled CRISPR-based​ screens can be greatly expanded by the application of single-cell​ sequencing to determine transcriptional activity and genetic perturbation in each individual cell192–196. Such an approach allows understanding of heterogeneity in a cell population as well as the elucidation of more complex mechanisms, beyond the viability readout that has historically been applied. This additional information is likely to be very useful for the prioritization of potential drug targets.

www.nature.com/nrd Reviews

of being well controlled as the only difference between analytically straightforward, a more efficient approach them is the removal of a single gene; however, this is a could be to use a ‘druggable genome’ library, effectively highly engineered system that may not recapitulate gene inserting that filter before the screen, and therefore function observed in tumours from patients. With a cell markedly reducing the resources and time needed to panel approach for novel target discovery in the context generate and analyse the data. On the basis of the his­ of a tumour suppressor loss, five to ten well-​matched torical success of target classes, sequence-​based and lines will often usually provide sufficient confidence to structure-​based prediction, catalytic activity, and capac­ support subsequent validation. Dependency surveys ity of binding of an endogenous ligand, some studies across the genetic spectrum of human cancer, such as have estimated that no more than a quarter of the human Project Achilles, Project DRIVE and Project Score, will, genome is druggable by conventional means (~5,000 however, require very large numbers of cell lines and are genes)131–133. These analyses suggest that using a drug­ possible only with significant resources. gable genome CRISPR library will eliminate more than Once the number of cell lines to be screened has been 70% of screen hits that are not conventionally druggable determined, gRNA library size is the primary determi­ when a whole-​genome library is used, allowing time and nant of the resources required. The potential advantage resources to remain focused on other critical activities of reducing library size without limiting actionable find­ relevant to target discovery, including screening in a ings can be appreciated when one considers the method broader panel of context-​specific cell lines and subse­ used for most large-​scale CRISPR-​based screens. These quent target validation strategies. Some studies have screens are usually conducted in a pooled format to successfully adopted such an approach by using a set of allow maximum throughput with multiple internal druggable genes defined as desired by the user132,134,135. controls. A library of gRNA constructs, usually pack­ One important consideration when one is constructing aged in lentivirus, is used to infect target cell lines and a druggable genome library is the value of adding a set the effect of each gene knockout on cell growth can be of additional, conventionally undruggable genes relevant individually assessed using next-​generation sequencing. to cancer biology, such as the MAPK signalling path­ After an appropriate interval (usually 1–3 weeks) the way and all known recurrently mutated genes in cancer. abundance of each gRNA is measured: guide sequences The undruggable genes add great value by placing novel associated with loss of cell viability will be depleted in targets identified by screening into known biological the postinfection pool compared with their abundance pathways. They also provide both validation for hits in the preinfection pool. As library size is not a technical and biological rationale when one is selecting poten­ limitation for pooled screening, exome-​wide interro­ tial drug targets for validation. Regardless of the defi­ gation is feasible for a limited number of cell lines with­ nition, druggability is in the eyes of the beholder, new out access to large-​format robotics. By contrast, for drug discovery paradigms such as induced proteolysis-​ plate-​based screens, a tenfold increase in gRNA library targeting chimaeras are emerging40,41 and the scope of size, such as that between a 500-gene kinome and a any druggable genome library will need to be revisited as 5000-gene druggable genome, requires sophisticated technology develops. automation equipment for a robust scale-​up. However, even for pooled screening, library size does define the Targeting tumour suppressor gene loss total number of cells required for each screen and can Tumour suppressor gene loss is a central mechanism by result in rate-​limiting amounts of tissue culture when which normal cells undergo malignant transformation, multiple cell line panels are screened with a large library often by causing or allowing genome instability. Many for pooled screens. Larger cell line panels are important tumour suppressor genes have been well characterized, to adequately power a comparison between wild-​type such as TP53, RB1 and BRCA1, but tumour suppressor and mutant cell lines and account for secondary genetic gene loss is by definition undruggable, as the function changes present in all cancer cells. Therefore, limiting and often the genes themselves are lost. Identification of the size of the gRNA library size does not limit target druggable synthetic lethal partners is currently the only discovery if the number of genes interrogated in a single way of targeting the functional loss of tumour suppressor screen is also limited at the same time. Moreover, in case genes in cancer. of fixed resourcing, limiting the size of the gRNA library As noted earlier, the discovery of PARP inhibition allows analysis of a larger number of cell lines with the and BRCA1 or BRCA2 mutation as a synthetic lethal same amount of tissue culture resource, which in turn interaction by both Ashworth and colleagues and adds statistical power to target discovery efforts. Helleday and colleagues in 2005 (refs75,79) is the first to be When the experimental goal is drug target discov­ translated into clinical benefit for patients. However, this ery, efficiency and productivity are important consider­ discovery was not made by genomic screening, rather it ations, and several options may be considered. Kinome was hypothesis-​driven76,79. It was also blessed by a bit of and phosphatome libraries have been used effectively good luck: a small-​molecule PARP inhibitor had already for this purpose, but although they are manageable in been developed with plans for use as a cytotoxic agent, size (518 and 298 genes respectively) many novel targets the role of BRCA1 and BRCA2 in DNA damage response will fall outside these target classes. Project Score effec­ was known, and the synthetic lethal effect of PARP inhib­ tively used a genome-​scale library to identify almost 300 ition with BRCA loss is very strong74. However, the ini­ putative novel druggable targets, but used a postscreen tial enthusiasm that additional hypothesis-​driven efforts filter to eliminate genes that are considered undruggable would identify druggable synthetic lethal partners for by conventional methods92. Although this approach is other DNA damage-​related genes has not yet been

Nature Reviews | Drug Discovery Reviews

Normal cell Tumour cell

Cell death Cell death

SAM-competitive PRMT5 inhibitors

SAM SAH SAM SAHA MTA MTA MTA Met MTA MTA MTA

MTA MTA PRMT5 PRRMMT5T

MTAP MTAP Me Substrate Substrate Me MTAP MTAP

MTA-cooperative PRMT5 inhibitors No effect Cell death

Fig. 2 | PrMT5 and MTAP are a synthetic lethal pair. Methylthioadenosine phosphorylase (MTAP) is frequently deleted in human tumours, causing its substrate methyl-5′-thioadenosine (MTA) to accumulate. MTA functions as a competitive inhibitor for the S-​methyl-5′-thioadenosine phosphorylase (PRMT5)-activating cofactor S-​adenosylmethionine (SAM), which provides a methyl donor group for client proteins. Therefore, MTA accumulation reduces, but does not eliminate, PRMT5 activity. PRMT5 inhibitors that leverage the accumulation of MTA (MTA-cooperative​ inhibitors) should drive selectivity for the MTAP-deleted​ tumours while sparing the wild-type​ normal cells. Existing PRMT5 inhibitors that are SAM cooperative do not leverage MTA accumulation, inhibit PRMT5 regardless of the genetic context and therefore are not selective for MTAP-deleted​ tumours. SAH, S-​adenosyl-l-​homocysteine.

fulfilled with the lack of success of these efforts. This occurs when a ‘passenger’ gene adjacent to a tumour is at least partly due to the lack of selective and potent suppressor gene is lost along with the ‘driver’ gene. In pharma­cologic reagents136. Furthermore, hypothesis-​ this case, MTAP is the passenger gene and is frequently driven discovery of synthetic lethal DNA repair path­ co-​deleted with the driver cyclin-​dependent kinase ways is limited by the sheer volume of hypotheses to test inhibitor (CDKN2A, encoding p16-INK4)72–74. Several with seven well-​described unique DNA repair pathways groups discovered that MTAP-​null cancer cells have a and multiple genes within each. marked dependency on PRMT5, an essential methyl­ Given the potentially large number of synthetic lethal transferase, making these cells much more suscepti­ interactions in the human genome, it is clear that dis­ ble to PRMT5 knockdown than those without MTAP covery of additional synthetic lethal pairs amenable to deletion72–74 (Fig. 2). This dependency occurs because drug discovery requires a functional genomic approach. MTAP-​null cells accumulate high levels of the PRMT5 To this end, a CRISPR library can be used to identify inhibitory cofactor S-​methyl-5′-thioadenosine (MTA). synthetic lethal ‘hits’ in cell line panels that share loss of As a result, PRMT5 is partially inhibited at the baseline a specific tumour suppressor gene matched as closely in MTAP-​null cells, and they are profoundly sensitive to as possibly to cell lines that retain wild-​type function of further reduction of PRMT5 activity (for example, by the gene of interest. Hits from these screens are poten­ genetic knockdown). This dependency provides the tial drug targets, and the patient population expected to potential for a large therapeutic window for PRMT5 benefit from inhibiting the novel targets is defined by the inhibitors in patients with MTAP-​deleted tumours given genetic context being interrogated. that normal cells (without MTAP deletion) would be largely spared, limiting toxicity. MTAP deletion and PRMT5 inhibition: synthetic lethal- MTAP is deleted in approximately 15% of all human ity beyond PARP inhibitors. One of the strongest and cancers, including more than 50% of glioblastomas most prevalent synthetic lethal interactions discov­ and 25% of pancreatic cancers; thus, development of ered by Project Achilles and Project DRIVE is PRMT5 an effective therapy for MTAP-​deleted tumours could dependence in cells with MTAP deletions72–74. This have high patient impact. However, the effect of PRMT5 dependency represents a subset of synthetic lethal­ knockdown has not been recapitulated with existing ity termed ‘collateral lethality’137. Collateral lethality PRMT5 inhibitors72–74. The lack of concordance between

www.nature.com/nrd Reviews

139–142 Competitive inhibitors genetic knockdown and pharmacologic inhibition in inhibition of CDK4 and CDK6 (CDK4/6) , an active Small molecules that compete this setting could suggest that functional genomics combination because of the crosstalk between ER and with the substrates or cofactors may not be a good drug target discovery tool, when it cell cycle signalling143. The discovery of this interaction when binding to the target is in fact due to the mechanism of action of existing came from profiling of a CDK4/6 inhibitor in a panel enzyme, resulting in functional inhibition. By contrast, an inhibitors. PRMT5 has two cofactors: the activating of 47 breast cancer cell lines and finding of significant uncompetitive inhibitor binds cofactor S-​adenosylmethionine (SAM) and the inhib­ growth inhibition clearly limited to the ER-​positive cell to an enzyme–substrate itory cofactor MTA. All known PRMT5 inhibitors are lines in the panel. CDK4/6 inhibition was therefore complex more tightly than to either SAM-​cooperative inhibitors (GlaxoSmithKline) tested in combination with several modulators of oes­ the enzyme alone, also or MTA-​competitive inhibitors (Eli Lilly and Company, trogen blockade, including tamoxifen, and was found to resulting in functional 139,142 inhibition. Johnson & Johnson). Because MTA accumulates be synergistic both preclinically and in patients . As in MTAP-​deleted cancers to much higher levels than in single agents, CDK4/6 inhibitors are not active in ER-​ normal cells, an MTA-​cooperative and SAM-​competitive positive breast cancer, and the combination of CDK4/6 PRMT5 inhibitor will be required to recapitulate the inhibition and oestrogen blockade is not active in ER-​ effect of PRMT5 knockdown73. Such an inhibitor would negative breast cancer subtypes. Thus, CDK4/CDK6 is increase the amount of inactive PRMT5 (bound to a context-​dependent ‘unmarked’ oncogene in the ER-​ MTA) relative to active PRMT5 (bound to SAM), which positive breast cancer context. This hypothesis-​driven would result in death of MTAP-​deficient cells but not approach led to an important clinical advance, but it is wild-​type cells. Existing PRMT5 inhibitors do not act by not scalable. Empiric approaches to combination discov­ this mechanism and therefore kill both MTAP-​deleted ery have been largely unsuccessful, and are prohibitively cells and wild-​type cells at similar exposures, resulting in expensive if performed in clinical trials as is common­ toxicity with limited activity73. MTA-​cooperative, SAM-​ place at present with immunologic agents, highlighting competitive PRMT5 inhibitors are likely in development the potential value of a functional genomics strategy. but not currently available. Instead, the first effort to exploit this synthetic lethal pair has been with MAT2A Novel drug combination discovery. One approach to dis­ inhibitors, another component of the metabolic path­ covering drug combinations is to uncover novel context-​ way that acts by reducing SAM levels. The first MAT2A dependent unmarked oncogenes using a CRISPR library inhibitor is now in clinical trials but limited efficacy data in combination with a targeted drug that is relevant to have been released138. a specific genetic context. We define this as an ‘anchor’ screen, in which a targeted drug is the ‘anchor’ (Fig. 3). Beyond synthetic lethal targets This approach was used by Bernards and colleagues in Combining CRISPR-​based screening with known tar- 2012 using shRNA-​based screening (before CRISPR sys­ geted drugs. DNA sequencing provided the means to tems were widely available) to identify combination drug identify the first wave of genetic drug targets for spe­ targets that would enhance the efficacy of the BRAF cific cancer subtypes because they are ‘marked’ with inhibitor vemurafenib in BRAF-​mutant colon cancer144. fixed genetic alterations, and many of these genetically Vemurafenib has a 50–60% overall response rate in mel­ altered oncogenes have now been successfully drugged. anoma19,20 but only a 5% overall response rate in colon As noted earlier, BRAF inhibitors and HER2 inhibitors cancers145,146 with activating BRAF mutations. With use are very active drugs in BRAF-​mutant melanoma and of a focused library of shRNAs representing 518 kinases ERBB2-amplified breast cancer, respectively. However, (the ‘kinome’), a total of six colon cancer cell lines with these inhibitors have limited activity, despite the pres­ or without BRAFV600E activating mutations were screened ence of the same genetic alterations, in colon and gastric for genetic dependencies in the presence of vemurafenib. cancer. Thus, BRAF and ERBB2 are context-​dependent This drug anchor screen identified EGFR as a combi­ ‘marked’ oncogenes. We define ‘unmarked oncogenes’ nation target with vemurafenib in this context144. This as genes that are not genetically activated through combination has since been shown to be clinically active, mutation, amplification or translocation but are sim­ with an EGFR inhibitor doubling the response rate of ilarly important oncogenic drivers in specific genetic BRAF inhibition alone in BRAF-​mutant colon can­ contexts. On the basis of the well-​described concept of cer, yet unfortunately remaining at a very modest 10% non-​oncogene addition, which is driven by ‘unmarked overall response rate147. The Bernards group also used a oncogenes’, we postulate that there are many such onco­ 298-gene ‘phosphatome’ shRNA library to identify genes relevant to specific genetic contexts Many of PTPN11 inhibitors as another way to enhance the effect these genes will be good drug targets but will require a of BRAF inhibitors in this setting148, and clinical trials to functional genomic screening approach for discovery, test this hypothesis are ongoing149,150. Garraway and col­ as hypothesis-​driven approaches are hampered by our leagues also conducted a BRAF inhibitor anchor screen limited understanding of the biology of complex systems in a BRAF-​mutant colon cancer cell line but instead used and the reality that empiric approaches are not scalable. a larger, genome-​scale shRNA library and discovered A modified synthetic lethal screening platform can be additional combination targets, although the conven­ used to identify novel context-​dependent unmarked tionally druggable targets identified by this much larger oncogenes that may be good drug targets for single-​agent screen were included in the smaller, more manageable or combination therapy. kinome and phosphatome libraries144,148,151. Finally, a This concept is exemplified by the finding that breast similar approach has been taken to identify combina­ cancers that express oestrogen receptor (ER positive) tion partners for MEK inhibitors as several potent and are sensitive to a combination of inhibition of ER and selective MEK inhibitors are in clinical use, but they

Nature Reviews | Drug Discovery Reviews

a

Drug NGS

CRISPR druggable genome library NGS No drug

Genes that sensitize cells to anchor drug are depleted

b 0

0.1

0.2

0.3

0.4

0.5

0.6 False discovery rate discovery False Log2 fold change drug versus plasmid versus fold change drug Log2 0.7 Log2 fold change untreated versus plasmid 0.8 Non-targeting sgRNA 0.9 Essential gene sgRNA Dropout gene sgRNA 1 Rescue gene sgRNA –0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1 Effect size Potential Potential drug targets resistance markers

Fig. 3 | identifying novel combination targets using crisPr screening. a | Relevant cancer cell lines with or without a mutation of interest are infected with a CRISPR library and then passaged in the presence or absence of selective small-molecule​ inhibitors at their clinically relevant doses. At the end of the treatment period, cells are harvested, and the abundance of each guide RNA (gRNA) is evaluated using next-generation​ sequencing (NGS). Levels of gRNA are compared between drug treatment samples and control (no drug) samples to identify those gRNAs that are selectively depleted in the presence of the drug (green gRNAs). b | gRNA counts from NGS are quantified, then normalized and compared with the plasmid library. Genes that sensitize (which are potential drug targets in combination with the anchor drug) or rescue (which are markers of resistance to the anchor drug) cell viability to the anchor drug would selectively be depleted or enriched in the drug-​treated group but not the control group (left panel). The results of this effect are statistically quantified using MAGeCK198 to identify both potential drug targets (dropouts) and resistance marker (rescue) genes (right panel). sgRNA, single-guide​ RNA.

are minimally active in KRAS-​mutant NSCLC152,153. An accumulation of genetic alterations in cancer cells cre­ shRNA-​based kinome anchor screen identified ERBB3 ates neoantigens, which should be recognized by the blockade as an approach for enhancing the effect of immune system as ‘foreign’ and trigger immune destruc­ MEK inhibition in this setting, a hypothesis that remains tion of nascent cancers154,155. However, all cancers that to be tested in clinical studies130. These anchor screens, become clinically relevant have, by definition, escaped although informative, can now be enhanced by use of immune destruction. A wide variety of immune evasion CRISPR technologies as the genetic screening tool. As a mechanisms have been postulated, including immune result, anchor screens can be widely applied to identify editing, T cell exhaustion and an inhibitory microenvi­ drug combinations with either known or novel targets ronment156, but important drivers of immune evasion that will be required to start seeing sustained, complete must emanate at least in part from the cancer cell itself. responses in patients with metastatic solid tumours. However, the genetics of tumour-​intrinsic immune eva­ sion have only recently begun to be described155,157,158 Screening for non-cell-autonomous​ targets: immune eva- and no drug targets with the potential to reverse this sion context. The classic definition of synthetic lethal­ hallmark of cancer have yet been discovered. ity is cell autonomous, but a synthetic lethal approach In 2016, Ribas and colleagues reported that loss-​ can be adapted to identify druggable targets that do of-function mutations in JAK1 enhance immune evasion not kill cancer cells directly, but rather attract immune and conferred anti-​PD1 resistance in a patient treated cells to destroy them. As has been well described, the with a checkpoint inhibitor158. This study provided the

www.nature.com/nrd Reviews

first evidence that a tumour cell can evade immune pres­ stimulating host immunity, such as checkpoint inhibi­ sure by evolving intrinsic genetic changes. Additional tors. In addition, single-​agent activity may be possible recurring loss-​of-function mutations were subsequently in cases in which active T cell infiltration has already identified in genes encoding other components of the occurred in the tumour environment and in which the antigen presentation machinery, such as components tumour cell-​driven immune evasion mechanism is of the class I major histocompatibility complex HLA the key gatekeeping event for preventing immune and B2M, further supporting this observation159–161. eradication. This approach has the potential to discover Emerging data also indicate that many well-​studied novel medicines for cancers that are insensitive to oncogenic drivers and tumour suppressor genes may immune checkpoint inhibition and have the advantage play prominent parts in immune evasion. One exam­ of defining a genetic patient selection strategy for clinical ple of a putative tumour-​intrinsic immune evasion development. gene is the protein kinase gene LKB1 (also known as Discovering tumour-​intrinsic immune evasion tar­ STK11), a tumour suppressor gene inactivated in ~20% gets requires a two-​step process: first, a genetic context of NSCLC97. Loss of function of LKB1 has been postu­ that confers immune evasion is identified and, second, lated to drive tumorigenesis through activation of the the drug targets that reverse such a phenotype are iden­ mechanistic target of rapamycin pathway in a cancer tified (Fig. 4). In the first step, a synthetic lethal-​based cell-​autonomous manner162. However, LKB1 loss also CRISPR screen can be applied by changing the readout results in accumulation of neutrophils with T cell sup­ from growth rate alterations or cell death to immune pressive effects and an increase in the levels of tumour-​ cell-​mediated lethality and using in vivo screening promoting cytokines163. A retrospective analysis of approaches. Providing a strong proof of principle, an tumours from patients who did not respond to treatment in vivo CRISPR screen that evaluated 2,700 candidate with PD1 inhibitors correlated LKB1 loss with reduced genes was recently reported in B16 syngeneic mice PD-​L1 expression164, further suggesting that LKB1 is a treated with PD1 checkpoint blockade169. When the bona fide suppressor of immune evasion, which may be tumours were collected and sequenced, genes driv­ one of its primary functions. Other genetic alterations ing immune evasion and immune sensitization were linked to immune evasion include MYCN amplifica­ identified by comparison of their relative abundance tion, which limits T cell infiltration in neuroblastoma in tumours grown under increasing immune pressure. through downregulation of the interferon response165, Known immune evasion genetic contexts, such as JAK1 CASP8 loss of function, which rescues cancer cells and B2M loss of function, scored highly in this study. from T cell-​mediated lethality by blocking the tumour However, to expand these findings, studies that com­ necrosis factor pathway166 and PTEN loss of function, prehensively identify the genetic contexts that drive which promotes resistance to T cell killing by increasing immune evasion are needed. These context discovery production of immunosuppressive cytokines167. Thus, a screens will define the genetic contexts in which to systematic functional genomic evaluation of all known conduct target discovery screens, the second step in cancer genes is warranted to then allow identification of tumour-​intrinsic immune evasion target discovery. drug targets that reverse immune evasion signalling. In Once an immune evasion genetic context has been this setting, the immune evasion gene (such as LKB1) discovered, target discovery screens can be performed provides the context in which a novel drug target (to be either in vivo or in vitro. In vivo models are more reflec­ determined) functions as the patient selection biomarker tive of the human immune system, whereas in vitro for clinical trial development. screens are more reflective of histology-​specific human Approved drugs that target the immune cell check­ cancer genetics. In vivo screens have the advantage of an points PD1, PD-​L1 and CTLA4 are now widely used intact immune system, but require special consideration. as anticancer therapy, but these medicines are directed There are limited numbers of syngeneic mouse models, towards and activate T cells as opposed to targeting and those that exist typically have chemically induced tumour-​intrinsic immune evasion mechanisms. Building tumours and therefore a mutation spectrum that is not on the success of the current treatment para­digm, the reflective of corresponding human cancers. In addition, discovery of novel immuno-​oncology targets and devel­ the throughput of in vivo CRISPR screening is limited opment efforts are almost exclusively focused on host by the library size (generally fewer than 4,000 sgRNAs) immunity, such as modulating T cells, natural killer and the number of cells that can be implanted in each cells, macrophages and the tumour micro­environment. mouse. By contrast, there are many genetically engi­ Therefore, a novel approach that targets tumour-​intrinsic neered murine models designed to study human cancer immune evasion mechanisms (likely in combination genetics, but they are mostly non-​immunogenic and with checkpoint inhibitors) will have several advan­ are therefore less useful for immune evasion context tages. First, it will open a largely unexplored target and target discovery. space. Second, targeting tumour-​intrinsic mechanisms When the mechanism of gene-​specific immune eva­ will direct immune cells to the tumour specifically and sion is known or postulated, in vitro screening with a may limit the systemic autoimmune toxicity that is the relevant immune readout such as PD-​L1 expression primary toxic effect of checkpoint inhibitors168. Finally, may be the most efficient approach for identifying drug the genetic context in which specific immune evasion targets. For example, JAK1 loss of function mediates targets are active will define patient selection biomarkers immune evasion through suppression of interferon that have been elusive for checkpoint inhibitor therapy. signalling170,171. Therefore, a target discovery screen in Such drugs could be used in combination with reagents human cancer cell lines that harbour a JAK1 inactivating

Nature Reviews | Drug Discovery Reviews

a Context discovery screen

Immuno- Increasing Syngenic tumour cells immune competent Genes that confer immune CRISPR context library (such + anti-PD1 pressure as tumour suppressor genes) evasion are enriched in immunocompetent mice b Target discovery screen Mutant Immuno- competent

CRISPR druggable Wild-type Genes that reverse context- genome library dependent immune evasion are depleted in mutant cells Immuno- but not wild-type cells compromised Syngenic cells with defined immune evasion context mutation

Fig. 4 | identifying synthetic lethal drug targets that reverse tumour-intrinsic​ immune evasion. a | Syngeneic mouse models with known responsiveness to immune therapies, such as MC38 and CT26, can be used to screen mice for genes that drive immune evasion when inactivated (immune evasion genes). A ‘context discovery’ CRISPR library that consists of candidate immune evasion genes can then be introduced into the tumour models as a pool using lentiviral infection. Infected cells are subsequently implanted into different mouse strains that are either immunocompromised or immunocompetent. Immune pressure in the immunocompetent mice can be further enhanced by activating T cells with an anti-programmed​ cell death 1 (anti-​PD1) antibody. Tumours are then allowed to grow , and at the time of tumour harvesting, cells with immune evasion gene loss will survive best in models with the highest immune pressure. Those genes can be identified using next-generation​ sequencing to measure the relative abundance of guide RNA.b | For the discovery of drug targets that reverse immune evasion driven in a specific genetic context, genetically engineered cell lines expressing genes that confer an immune evasion genetic context (mutant) and wild-type​ cells are infected with a druggable genome CRISPR library using the schema described above to supply immune pressure. Novel drug targets can be identified by identifying those guide RNAs depleted in the presence of immune pressure in mutant cells but not wild-type​ cells.

mutation could measure PD-​L1 induction following CRISPR-​based screen in a large panel of cancer cell interferon stimulation. Additional in vitro screen­ lines provides direct evidence that this number is likely ing approaches may include tumour cell and T cell in the hundreds, but the large majority will be context co-​cultures, such as the ovalbumin mouse model, in specific92. Applying the genetic concept of synthetic which T cells are engineered to express the relevant lethality with both its classic definition and with some T cell receptor for ovalbumin recognition, and the conceptual modifications to identify these drug tar­ tumour cells are modified to express chicken ovalbumin gets, and the genetic contexts (and therefore patients) antigen166. Such systems can be used to interrogate the in which they will be effective therapies, represents a phenotype of T cell-​mediated killing directly and could transformative opportunity for patients, and the tools bridge in vivo and in vitro approaches. with which to do this are becoming ever more powerful. As the underlying cancer genetics that drive immune Continuously evolving CRISPR technology, including evasion become better understood, the tools of synthetic single-​cell techniques, holds the promise of addressing lethal-​based target discovery can be used to discover novel the tumour heterogeneity involved in primary drug tumour-​intrinsic drug targets that reverse immune eva­ resistance and the genetic evolution that drives sec­ sion. Context-​dependent, tumour-​intrinsic, immune ondary resistance. Combination CRISPR vectors will evasion target discovery screens can be conducted using simplify the search for effective, context-​specific drug paired isogenic models engineered with the immune eva­ combinations, and base-​editing techniques are stream­ sion gene loss (such as JAK1 loss of function) and a wild-​ lining the construction of the tools needed for target type control. Potential drug targets are those genes that validation. Finally, the continued refinement of technol­ reverse the immune evasion phenotype when knocked ogies that will make in vivo CRISPR screening as scal­ out. These targets have the potential to overcome immune able as in vitro approaches is enabling the integration of checkpoint inhibitor resistance and thereby address a cancer genetics and immuno-​oncology and driving our significant unmet medical need. understanding of the genetic basis of immune evasion. Enhanced by these new tools, CRISPR-based functional Future directions genomic screening can be applied in increasingly unique We are reaching the limits of sequence-​based genetic ways to address the loss of tumour suppressor genes, target discovery for cancer but there are compelling unmarked oncogenes and non-cell-autonomous path­ reasons to believe that many cancer drug targets are ways, and will yield the next wave of effective targeted still to be identified. Recent analysis of a genome-​scale therapies.

www.nature.com/nrd Reviews

New technologies that expand the number of tar­ further compounding the time needed to bring truly gets that are considered druggable have the potential to transformative combinations to a wide spectrum of have a great impact on drug discovery. Targeted protein patients. Finally, evolving drug resistance is inevitable degradation is one such example that could markedly until complete tumour ablation is achieved, likely requir­ expand the number of druggable targets. Despite the ing a combination of targeted therapy and immunother­ great promise of these approaches, there remain many apy. Thus, understanding resistance mechanisms and the challenges in moving from the novel target discovery application of this approach in those settings will also be that is now within our grasp to medicines that are clin­ essential. Nonetheless, CRISPR, with all of its current ically effective. The large majority of novel targets will and future applications, is a tool that has opened the be intracellular proteins that require a small-​molecule door to an enormous range of possibilities. Many people inhibitor for clinical activity. Under the best of circum­ with cancer have been cured in our lifetimes, an achieve­ stances, 5 years from target discovery to a clinical proof ment not thought possible when synthetic lethality was of concept is considered ‘fast’, so time alone is a hurdle first discovered in fruit flies, and many more people that is difficult to avoid with current drug discovery will be cured as a result of the work now made possi­ technology. In addition to the frustratingly long time­ ble by applying the concept of synthetic lethality with lines for drug discovery, the impact of context depen­ the genome editing power of CRISPR technology. dency means that focused functional genomic discovery Published online xx xx xxxx approaches need to be applied to many cancer subsets,

1. Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: 19. Chapman, P. B. et al. Improved survival with 35. Shaw, A. T. & Engelman, J. A. Ceritinib in ALK-​ the next generation. Cell 144, 646–674 (2011). vemurafenib in melanoma with BRAF V600E rearranged non-small-cell​ lung cancer. N. Engl. J. Med. 2. Lander, E. S. et al. Initial sequencing and analysis mutation. N. Engl. J. Med. 364, 2507–2516 370, 2537–2539 (2014). of the human genome. Nature 409, 860–921 (2011). 36. Sawyers, C. Targeted cancer therapy. Nature 432, (2001). 20. Hauschild, A. et al. Dabrafenib in BRAF-​mutated 294–297 (2004). 3. Venter, J. C. et al. The sequence of the human metastatic melanoma: a multicentre, open-​label, 37. Lawrence, M. S. et al. Discovery and saturation genome. Science 291, 1304–1351 (2001). phase 3 randomised controlled trial. Lancet 380, analysis of cancer genes across 21 tumour types. 4. Hoadley, K. A. et al. Cell-​of-origin patterns dominate 358–365 (2012). Nature 505, 495–501 (2014). the molecular classification of 10,000 tumors from 21. Larkin, J. et al. Combined vemurafenib and This landmark study integrates large-​scale next-​ 33 types of cancer. Cell 173, 291–304.e6 (2018). cobimetinib in BRAF-​mutated melanoma. N. Engl. generation whole-exome​ sequencing data from 5. Tate, J. G. et al. COSMIC: the catalogue of somatic J. Med. 371, 1867–1876 (2014). human tumours to define the landscape of tumour mutations in cancer. Nucleic Acids Res. 47, 22. Robert, C. et al. Improved overall survival in suppressor genes and oncogenes across 21 D941–D947 (2019). melanoma with combined dabrafenib and trametinib. different cancer types. 6. International Cancer Genome Consortium. N. Engl. J. Med. 372, 30–39 (2015). 38. Bailey, M. H. et al. Comprehensive characterization International network of cancer genome projects. 23. Dummer, R. et al. Overall survival in patients with of cancer driver genes and mutations. Cell 173, Nature 464, 993–998 (2010). BRAF-​mutant melanoma receiving encorafenib 371–385 e318 (2018). 7. Cancer Genome Atlas Research Network et al. plus binimetinib versus vemurafenib or encorafenib 39. Beroukhim, R. et al. The landscape of somatic copy-​ The Cancer Genome Atlas Pan-​Cancer analysis project. (COLUMBUS): a multicentre, open-​label, randomised, number alteration across human cancers. Nature Nat. Genet. 45, 1113–1120 (2013). phase 3 trial. Lancet Oncol. 19, 1315–1327 463, 899–905 (2010). 8. Hochhaus, A. et al. Long-​term outcomes of imatinib (2018). 40. Churcher, I. Protac-​induced protein degradation in treatment for chronic myeloid leukemia. N. Engl. J. Med. 24. Slamon, D. J. et al. Use of chemotherapy plus a drug discovery: breaking the rules or just making new 376, 917–927 (2017). monoclonal antibody against HER2 for metastatic ones? J. Med. Chem. 61, 444–452 (2018). 9. Druker, B. J. et al. Five-​year follow-​up of patients breast cancer that overexpresses HER2. N. Engl. J. Med. 41. Lai, A. C. & Crews, C. M. Induced protein degradation: receiving imatinib for chronic myeloid leukemia. 344, 783–792 (2001). an emerging drug discovery paradigm. Nat. Rev. Drug N. Engl. J. Med. 355, 2408–2417 (2006). 25. Baselga, J. et al. Pertuzumab plus trastuzumab Discov. 16, 101–114 (2017). 10. Kantarjian, H. et al. Improved survival in chronic plus docetaxel for metastatic breast cancer. N. Engl. 42. Robert, C. et al. Nivolumab in previously untreated myeloid leukemia since the introduction of imatinib J. Med. 366, 109–119 (2012). melanoma without BRAF mutation. N. Engl. J. Med. therapy: a single-​institution historical experience. 26. Mok, T. S. et al. Gefitinib or carboplatin-​paclitaxel in 372, 320–330 (2015). Blood 119, 1981–1987 (2012). pulmonary adenocarcinoma. N. Engl. J. Med. 361, 43. Schachter, J. et al. Pembrolizumab versus ipilimumab 11. Roy, L. et al. Survival advantage from imatinib 947–957 (2009). for advanced melanoma: final overall survival results of compared with the combination interferon-​α plus 27. Maemondo, M. et al. Gefitinib or chemotherapy a multicentre, randomised, open-​label phase 3 study cytarabine in chronic-​phase chronic myelogenous for non-​small-cell lung cancer with mutated EGFR. (KEYNOTE-006). Lancet 390, 1853–1862 (2017). leukemia: historical comparison between two phase 3 N. Engl. J. Med. 362, 2380–2388 (2010). 44. Weber, J. S. et al. Nivolumab versus chemotherapy in trials. Blood 108, 1478–1484 (2006). 28. Mitsudomi, T. et al. Gefitinib versus cisplatin plus patients with advanced melanoma who progressed 12. Verweij, J. et al. Progression-​free survival in docetaxel in patients with non-​small-cell lung cancer after anti-​CTLA-4 treatment (CheckMate 037): gastrointestinal stromal tumours with high-​dose harbouring mutations of the epidermal growth factor a randomised, controlled, open-​label, phase 3 trial. imatinib: randomised trial. Lancet 364, 1127–1134 receptor (WJTOG3405): an open label, randomised Lancet Oncol. 16, 375–384 (2015). (2004). phase 3 trial. Lancet Oncol. 11, 121–128 (2010). 45. Motzer, R. J. et al. Nivolumab versus everolimus in 13. Blanke, C. D. et al. Phase III randomized, intergroup 29. Zhou, C. et al. Erlotinib versus chemotherapy as advanced renal-cell​ carcinoma. N. Engl. J. Med. 373, trial assessing imatinib mesylate at two dose levels first-​line treatment for patients with advanced EGFR 1803–1813 (2015). in patients with unresectable or metastatic mutation-​positive non-​small-cell lung cancer 46. Brahmer, J. et al. Nivolumab versus docetaxel in gastrointestinal stromal tumors expressing the kit (OPTIMAL, CTONG-0802): a multicentre, open-​label, advanced squamous-cell​ non-small-cell​ lung cancer. receptor tyrosine kinase: S0033. J. Clin. Oncol. 26, randomised, phase 3 study. Lancet Oncol. 12, N. Engl. J. Med. 373, 123–135 (2015). 626–632 (2008). 735–742 (2011). 47. Borghaei, H. et al. Nivolumab versus docetaxel in 14. Blanke, C. D. et al. Long-​term results from a 30. Rosell, R. et al. Erlotinib versus standard advanced nonsquamous non-​small-cell lung cancer. randomized phase II trial of standard- versus higher-​ chemotherapy as first-​line treatment for European N. Engl. J. Med. 373, 1627–1639 (2015). dose imatinib mesylate for patients with unresectable patients with advanced EGFR mutation-​positive non-​ 48. Reck, M. et al. Pembrolizumab versus chemotherapy or metastatic gastrointestinal stromal tumors small-cell lung cancer (EURTAC): a multicentre, open-​ for PD-​L1-positive non-​small-cell lung cancer. N. Engl. expressing KIT. J. Clin. Oncol. 26, 620–625 (2008). label, randomised phase 3 trial. Lancet Oncol. 13, J. Med. 375, 1823–1833 (2016). 15. Demetri, G. D. et al. Efficacy and safety of imatinib 239–246 (2012). 49. Langer, C. J. et al. Carboplatin and pemetrexed with mesylate in advanced gastrointestinal stromal tumors. 31. Sequist, L. V. et al. Phase III study of afatinib or or without pembrolizumab for advanced, non-​ N. Engl. J. Med. 347, 472–480 (2002). cisplatin plus pemetrexed in patients with metastatic squamous non-​small-cell lung cancer: a randomised, 16. Bollag, G. et al. Clinical efficacy of a RAF inhibitor lung adenocarcinoma with EGFR mutations. J. Clin. phase 2 cohort of the open-​label KEYNOTE-021 study. needs broad target blockade in BRAF-​mutant Oncol. 31, 3327–3334 (2013). Lancet Oncol. 17, 1497–1508 (2016). melanoma. Nature 467, 596–599 (2010). 32. Soria, J. C. et al. Osimertinib in untreated EGFR-​ 50. Sun, C., Mezzadra, R. & Schumacher, T. N. Regulation 17. Rheault, T. R. et al. Discovery of dabrafenib: mutated advanced non-​small-cell lung cancer. N. Engl. and function of the PD-​L1 checkpoint. Immunity 48, a selective inhibitor of Raf kinases with antitumor J. Med. 378, 113–125 (2018). 434–452 (2018). activity against B-​Raf-driven tumors. ACS Med. 33. Solomon, B. J. et al. First-​line crizotinib versus 51. Hodi, F. S. et al. Improved survival with ipilimumab in Chem. Lett. 4, 358–362 (2013). chemotherapy in ALK-​positive lung cancer. N. Engl. patients with metastatic melanoma. N. Engl. J. Med. 18. Delord, J. P. et al. Phase I dose-​escalation and J. Med. 371, 2167–2177 (2014). 363, 711–723 (2010). -expansion study of the BRAF inhibitor encorafenib 34. Shaw, A. T. et al. Crizotinib versus chemotherapy in 52. Larkin, J. et al. Combined nivolumab and ipilimumab (LGX818) in metastatic BRAF-​mutant melanoma. advanced ALK-​positive lung cancer. N. Engl. J. Med. or monotherapy in untreated melanoma. N. Engl. Clin. Cancer Res. 23, 5339–5348 (2017). 368, 2385–2394 (2013). J. Med. 373, 23–34 (2015).

Nature Reviews | Drug Discovery Reviews

53. Tang, J. et al. The clinical trial landscape for PD1/ 77. Ashworth, A. A synthetic lethal therapeutic approach: for the survival of cancer cells with microsatellite PDL1 immune checkpoint inhibitors. Nat. Rev. Drug poly(ADP) ribose polymerase inhibitors for the instability. iScience 13, 488–497 (2019). Discov. 17, 854–855 (2018). treatment of cancers deficient in DNA double-​strand 96. Gillison, M. L., Chaturvedi, A. K., Anderson, W. F. 54. Weinstein, I. B. Addiction to oncogenes–the Achilles break repair. J. Clin. Oncol. 26, 3785–3790 (2008). & Fakhry, C. Epidemiology of human papillomavirus-​ heal of cancer. Science 297, 63–64 (2002). 78. Fong, P. C. et al. Inhibition of poly(ADP-​ribose) positive head and neck squamous cell carcinoma. 55. Solimini, N. L., Luo, J. & Elledge, S. J. Non-​oncogene polymerase in tumors from BRCA mutation carriers. J. Clin. Oncol. 33, 3235–3242 (2015). addiction and the stress phenotype of cancer cells. N. Engl. J. Med. 361, 123–134 (2009). 97. Cancer Genome Atlas Research Network. Cell 130, 986–988 (2007). 79. Bryant, H. E. et al. Specific killing of BRCA2-deficient Comprehensive molecular profiling of lung 56. Dobzhansky, T. Genetics of natural populations; tumours with inhibitors of poly(ADP-​ribose) adenocarcinoma. Nature 511, 543–550 (2014). recombination and variability in populations of polymerase. Nature 434, 913–917 (2005). 98. Kwak, E. L. et al. Anaplastic lymphoma kinase Drosophila pseudoobscura. Genetics 31, 269–290 This article describes the discovery of the PARP– inhibition in non-​small-cell lung cancer. N. Engl. J. Med. (1946). BRCA2 synthetic lethal relationship using a 363, 1693–1703 (2010). 57. Hartwell, L. H., Szankasi, P., Roberts, C. J., Murray, A. W. pharmacologic PARP inhibitor and isogenic mouse 99. Price, C. et al. Genome-​wide interrogation of human & Friend, S. H. Integrating genetic approaches into embryo fibrobast cell lines. cancers identifies EGLN1 dependency in clear cell the discovery of anticancer drugs. Science 278, 80. Murai, J. et al. Trapping of PARP1 and PARP2 by ovarian cancers. Cancer Res. 79, 2564–2579 (2019). 1064–1068 (1997). clinical PARP inhibitors. Cancer Res. 72, 5588–5599 100. Kaelin, W. G. Jr & Ratcliffe, P. J. Oxygen sensing by These authors proposed, for the first time, (2012). metazoans: the central role of the HIF hydroxylase applying the genetic principle of synthetic lethality 81. Hopkins, T. A. et al. Mechanistic dissection of PARP1 pathway. Mol. Cell 30, 393–402 (2008). using yeast genetic screens to identify novel trapping and the impact on in vivo tolerability and 101. Briggs, K. J., Min, C., Zhang, H. & Huang, A. Abstract vulnerabilities in cancer cells based on defined efficacy of PARP inhibitors. Mol. Cancer Res. 13, 2892: EGLN1 is a synthetic lethal target in ARID1A-​ genetic defects. 1465–1477 (2015). mutant ovarian cancer. J. Cancer Res. 78, 2892–2892 58. Kaelin, W. G. Jr Choosing anticancer drug targets in 82. Hopkins, T. A. et al. PARP1 Trapping by PARP inhibitors (2018). the postgenomic era. J. Clin. Invest. 104, 1503–1506 drives cytotoxicity in both cancer cells and healthy 102. Jones, S. et al. Frequent mutations of chromatin (1999). bone marrow. Mol. Cancer Res. 17, 409–419 (2019). remodeling gene ARID1A in ovarian clear cell 59. Kaelin, W. G. Jr The concept of synthetic lethality in 83. Smith, I. et al. Evaluation of RNAi and CRISPR carcinoma. Science 330, 228–231 (2010). the context of anticancer therapy. Nat. Rev. Cancer 5, technologies by large-​scale gene expression profiling 103. Wiegand, K. C. et al. ARID1A mutations in 689–698 (2005). in the connectivity map. PLOS Biol. 15, e2003213 endometriosis-associated​ ovarian carcinomas. 60. Sturtevant, A. H. A highly specific complementary (2017). N. Engl. J. Med. 363, 1532–1543 (2010). lethal system in Drosophila melanogaster. Genetics 84. Morgens, D. W., Deans, R. M., Li, A. & Bassik, M. C. 104. Arana, M. E. et al. synthesis by human DNA polymerase 41, 118–123 (1956). Systematic comparison of CRISPR/Cas9 and RNAi theta. Nucleic Acids Res. 36, 3847–3856 (2008). 61. Lucchesi, J. C. Synthetic lethality and semi-​lethality screens for essential genes. Nat. Biotechnol. 4, 105. Chan, S. H., Yu, A. M. & McVey, M. Dual roles for DNA among functionally related mutants of Drosophila 634–636 (2016). polymerase theta in alternative end-​joining repair of melanogaster. Genet. 59, 37–44 (1968). 85. Evers, B. et al. CRISPR knockout screening outperforms double-strand​ breaks in Drosophila. PLOS Genet. 6, 62. Bridges, C. B. The origin of variations in sexual and shRNA and CRISPRi in identifying essential genes. e1001005 (2010). sex-limited​ characters. Am. Nat. 56, 51–63 (1922). Nat. Biotechnol. 34, 631–633 (2016). 106. Roerink, S. F., van Schendel, R. & Tijsterman, M. 63. Bender, A. & Pringle, J. R. Use of a screen for synthetic 86. Barretina, J. et al. The Cancer Cell Line Encyclopedia Polymerase theta-mediated​ end joining of replication-​ lethal and multicopy suppressee mutants to identify enables predictive modelling of anticancer drug associated DNA breaks in C. elegans. Genome Res. two new genes involved in morphogenesis in sensitivity. Nature 483, 603–607 (2012). 24, 954–962 (2014). Saccharomyces cerevisiae. Mol. Cell Biol. 11, 87. McDonald, E. R. 3rd et al. Project DRIVE: a 107. Kent, T., Chandramouly, G., McDevitt, S. M., 1295–1305 (1991). compendium of cancer dependencies and synthetic Ozdemir, A. Y. & Pomerantz, R. T. Mechanism of 64. Kaiser, C. A. & Schekman, R. Distinct sets of SEC lethal relationships uncovered by large-​scale, deep microhomology-​mediated end-joining​ promoted by genes govern transport vesicle formation and fusion RNAi screening. Cell 170, 577–592.e10 (2017). human DNA polymerase theta. Nat. Struct. Mol. Biol. early in the secretory pathway. Cell 61, 723–733 This article from Novartis describes the results of a 22, 230–237 (2015). (1990). large-scale​ shRNA-​based functional genomic screen 108. Mateos-Gomez,​ P. A. et al. Mammalian polymerase 65. Hennessy, K. M., Lee, A., Chen, E. & Botstein, D. in an extensive panel of human cancer cell lines. theta promotes alternative NHEJ and suppresses A group of interacting yeast DNA replication genes. This group used the integrated genomic and recombination. Nature 518, 254–257 (2015). Genes Dev. 5, 958–969 (1991). pathway data to systemically interrogate synthetic 109. Ceccaldi, R. et al. Homologous-​recombination- 66. Mullenders, J. & Bernards, R. Loss-​of-function genetic lethal and other dependency relationships across deficient tumours are dependent on Poltheta-​ screens as a tool to improve the diagnosis and treatment multiple histologies. mediated repair. Nature 518, 258–262 (2015). of cancer. Oncogene 28, 4409–4420 (2009). 88. Cowley, G. S. et al. Parallel genome-​scale loss of 110. Mengwasser, K. E. et al. Genetic screens reveal 67. Bommi-Reddy,​ A. et al. Kinase requirements in human function screens in 216 cancer cell lines for the FEN1 and APEX2 as BRCA2 synthetic lethal targets. cells: III. Altered kinase requirements in VHL−/− cancer identification of context-​specific genetic dependencies. Mol. Cell 73, 885–899.e6 (2019). cells detected in a pilot synthetic lethal screen. Proc. Sci. Data 1, 140035 (2014). 111. Nichols, R. J. et al. RAS nucleotide cycling underlies Natl Acad. Sci. USA 105, 16484–16489 (2008). 89. Tsherniak, A. et al. Defining a cancer dependency the SHP2 phosphatase dependence of mutant BRAF-, 68. Hoffman, G. R. et al. Functional epigenetics approach map. Cell 170, 564–576.e16 (2017). NF1- and RAS-​driven cancers. Nat. Cell Biol. 20, identifies BRM/SMARCA2 as a critical synthetic lethal This article reports the results of a second 1064–1073 (2018). target in BRG1-deficient cancers. Proc. Natl Acad. Sci. shRNA-based functional genomic screen designed 112. Chen, Y. N. et al. Allosteric inhibition of SHP2 USA 111, 3128–3133 (2014). to identify genetic dependencies in approximately phosphatase inhibits cancers driven by receptor This is the first report to demonstrate the synthetic 500 human cancer cell lines. An informatics tool tyrosine kinases. Nature 535, 148–152 (2016). lethal interaction between SMARCA2 and SMARCA4. developed to eliminate false positive data from seed 113. Jinek, M. et al. A programmable dual-​RNA-guided 69. Wilson, B. G. et al. Residual complexes containing effects — and therefore to reduce the impact of DNA endonuclease in adaptive bacterial immunity. SMARCA2 (BRM) underlie the oncogenic drive of off-target​ gene silencing — is included in this report. Science 337, 816–821 (2012). SMARCA4 (BRG1) mutation. Mol. Cell Biol. 34, 90. Meyers, R. M. et al. Computational correction of copy 114. Cong, L. et al. Multiplex genome engineering using 1136–1144 (2014). number effect improves specificity of CRISPR-​Cas9 CRISPR/Cas systems. Science 339, 819–823 (2013). 70. Oike, T. et al. A synthetic lethality-​based strategy to essentiality screens in cancer cells. Nat. Genet. 49, 115. Mali, P. et al. RNA-​guided human genome engineering treat cancers harboring a genetic deficiency in the 1779–1784 (2017). via Cas9. Science 339, 823–826 (2013). chromatin remodeling factor BRG1. Cancer Res. 73, 91. Aguirre, A. J. et al. Genomic copy number dictates 116. Jinek, M. et al. RNA-​programmed genome editing in 5508–5518 (2013). a gene-independent​ cell response to CRISPR/Cas9 human cells. eLife 2, e00471 (2013). 71. Helming, K. C. et al. ARID1B is a specific vulnerability targeting. Cancer Discov. 6, 914–929 (2016). 117. Grimm, S. The art and design of genetic screens: in ARID1A-​mutant cancers. Nat. Med. 20, 251–254 92. Behan, F. M. et al. Prioritization of cancer therapeutic mammalian culture cells. Nat. Rev. Genet. 5, 179–189 (2014). targets using CRISPR-​Cas9 screens. Nature 568, (2004). 72. Kryukov, G. V. et al. MTAP deletion confers enhanced 511–516 (2019). 118. Boutros, M. & Ahringer, J. The art and design of dependency on the PRMT5 arginine methyltransferase This report from the Sanger Institute describes genetic screens: RNA interference. Nat. Rev. Genet. 9, in cancer cells. Science 351, 1214–1218 (2016). the results of a genome-wide​ CRISPR-based​ screen 554–566 (2008). 73. Mavrakis, K. J. et al. Disordered methionine in a diverse human cancer cell panel. The analysis 119. Birmingham, A. et al. 3’ UTR seed matches, but not metabolism in MTAP/CDKN2A-​deleted cancers leads of the data was specifically designed to identify overall identity, are associated with RNAi off-​targets. to dependence on PRMT5. Science 351, 1208–1213 potential therapeutic targets and associated Nat. Methods 3, 199–204 (2006). (2016). genetic alterations that could be used for patient 120. Buehler, E. et al. siRNA off-​target effects in genome-​ References 72 and 73 are the first to demonstrate selection in clinical studies. wide screens identify signaling pathway members. the synthetic lethal interaction between PRMT5 93. Chan, E. M. et al. WRN helicase is a synthetic lethal Sci. Rep. 2, 428 (2012). and MTAP deletion. MTAP is lost as a passenger target in microsatellite unstable cancers. Nature 568, 121. Jackson, A. L. & Linsley, P. S. Noise amidst the silence: owing to proximity to CDKN2A; therefore, this is 551–556 (2019). off-target​ effects of siRNAs? Trends Genet. 20, 521–524 one of the first examples of collateral lethality. This article describes the discovery that the (2004). 74. Marjon, K. et al. MTAP deletions in cancer create helicase WRN has a synthetic lethal interaction 122. Kampmann, M. CRISPRi and CRISPRa screens in vulnerability to targeting of the MAT2A/PRMT5/ with microsatellite instability in colon cancer. mammalian cells for precision biology and medicine. RIOK1 axis. Cell Rep. 15, 574–587 (2016). Additional data suggest this effect will be histology ACS Chem. Biol. 13, 406–416 (2018). 75. Farmer, H. et al. Targeting the DNA repair defect in agnostic. 123. Shalem, O. et al. Genome-​scale CRISPR-​Cas9 BRCA mutant cells as a therapeutic strategy. Nature 94. Lieb, S. et al. Werner syndrome helicase is a selective knockout screening in human cells. Science 343, 434, 917–921 (2005). vulnerability of microsatellite instability-​high tumor 84–87 (2014). 76. Lord, C. J. & Ashworth, A. PARP inhibitors: synthetic cells. eLife 8, e43333 (2019). 124. Wang, T., Wei, J. J., Sabatini, D. M. & Lander, E. S. lethality in the clinic. Science 355, 1152–1158 95. Kategaya, L., Perumal, S. K., Hager, J. H. & Genetic screens in human cells using the CRISPR-​Cas9 (2017). Belmont, L. D. Werner syndrome helicase is required system. Science 343, 80–84 (2014).

www.nature.com/nrd Reviews

125. Koonin, E. V., Makarova, K. S. & Zhang, F. Diversity, 149. US National Library of Medicine. ClinicalTrials.gov 169. Manguso, R. T. et al. In vivo CRISPR screening classification and evolution of CRISPR-​Cas systems. https://clinicaltrials.gov/ct2/show/NCT03114319 identifies Ptpn2 as a cancer immunotherapy target. Curr. Opin. Microbiol. 37, 67–78 (2017). (2019). Nature 547, 413–418 (2017). 126. Shmakov, S. et al. Diversity and evolution of class 2 150. US National Library of Medicine. ClinicalTrials.gov This study demonstrates the utility of in vivo CRISPR-Cas​ systems. Nat. Rev. Microbiol. 15, https://clinicaltrials.gov/ct2/show/NCT03634982 CRISPR genomic screening in a syngeneic mouse 169–182 (2017). (2019). model to identify tumour-intrinsic​ critical nodes of 127. Ishino, Y., Krupovic, M. & Forterre, P. History of 151. Whittaker, S. R. et al. Combined pan-​RAF and MEK immune modulation in solid tumours, and describes CRISPR-​Cas from encounter with a mysterious inhibition overcomes multiple resistance mechanisms several specific genes that could be therapeutic repeated sequence to genome editing technology. to selective RAF inhibitors. Mol. Cancer Ther. 14, targets designed to reverse the immune evasion J. Bacteriol. 200, e00580-17 (2018). 2700–2711 (2015). phenotype. 128. Zetsche, B. et al. Cpf1 is a single RNA-​guided 152. Blumenschein, G. R. Jr et al. A randomized phase II 170. Gao, J. et al. Loss of IFN-​gamma pathway genes endonuclease of a class 2 CRISPR-​Cas system. study of the MEK1/MEK2 inhibitor trametinib in tumor cells as a mechanism of resistance to Cell 163, 759–771 (2015). (GSK1120212) compared with docetaxel in KRAS-​ anti-CTLA-4​ therapy. Cell 167, 397–404 e399 These authors discover and develop the Cpf1 mutant advanced non-​small-cell lung cancer (NSCLC). (2016). system as an alternative gene-​editing approach Ann. Oncol. 26, 894–901 (2015). 171. Garcia-​Diaz, A. et al. Interferon receptor signaling to CRISPR–Cas9 and demonstrate the enhanced 153. Janne, P. A. et al. Selumetinib plus docetaxel for pathways regulating PD-​L1 and PD-​L2 expression. functionality of this system in combo-CRISPR​ KRAS-mutant​ advanced non-​small-cell lung cancer: Cell Rep. 19, 1189–1201 (2017). screens using this system due to a markedly a randomised, multicentre, placebo-​controlled, 172. Gilbert, L. A. et al. CRISPR-​mediated modular reduced recombination frequency. phase 2 study. Lancet Oncol. 14, 38–47 (2013). RNA-guided​ regulation of transcription in eukaryotes. 129. Fonfara, I., Richter, H., Bratovic, M., Le Rhun, A. & 154. Hacohen, N., Fritsch, E. F., Carter, T. A., Lander, E. S. Cell 154, 442–451 (2013). Charpentier, E. The CRISPR-​associated DNA-cleaving​ & Wu, C. J. Getting personal with neoantigen-​based 173. Gilbert, L. A. et al. Genome-​scale CRISPR-​mediated enzyme Cpf1 also processes precursor CRISPR RNA. therapeutic cancer vaccines. Cancer Immunol. Res. 1, control of gene repression and activation. Cell 159, Nature 532, 517–521 (2016). 11–15 (2013). 647–661 (2014). 130. Sun, C. et al. Intrinsic resistance to MEK inhibition 155. Schumacher, T. N. & Schreiber, R. D. Neoantigens These authors discover and describe the utility of in KRAS mutant lung and colon cancer through in cancer immunotherapy. Science 348, 69–74 CRISPR-​based gene silencing as an alternative to transcriptional induction of ERBB3. Cell Rep. 7, (2015). gene editing (‘cutting’) in genome-wide​ functional 86–93 (2014). 156. Vinay, D. S. et al. Immune evasion in cancer: screens. 131. Hopkins, A. L. & Groom, C. R. The druggable genome. mechanistic basis and therapeutic strategies. 174. Qi, L. S. et al. Repurposing CRISPR as an RNA-​guided Nat. Rev. Drug Discov. 1, 727–730 (2002). Semin. Cancer Biol. 35, S185–S198 (2015). platform for sequence-​specific control of gene 132. Griffith, M. et al. DGIdb: mining the druggable 157. Wellenstein, M. D. & de Visser, K. E. Cancer-​cell- expression. Cell 152, 1173–1183 (2013). genome. Nat. Methods 10, 1209–1210 (2013). intrinsic mechanisms shaping the tumor immune 175. Zetsche, B. et al. Multiplex gene editing by CRISPR-​ 133. Santos, R. et al. A comprehensive map of molecular landscape. Immunity 48, 399–416 (2018). Cpf1 using a single crRNA array. Nat. Biotechnol. 35, drug targets. Nat. Rev. Drug Discov. 16, 19–34 158. Zaretsky, J. M. et al. Mutations associated with 31–34 (2017). (2016). acquired resistance to PD-1 blockade in melanoma. 176. Horlbeck, M. A. et al. Mapping the genetic landscape 134. Duggan, S. P. et al. siRNA library screening identifies N. Engl. J. Med. 375, 819–829 (2016). of human cells. Cell 174, 953–967 e922 (2018). a druggable immune-​signature driving esophageal This group discovered and described an 177. Du, D. et al. Genetic interaction mapping in adenocarcinoma cell growth. Cell Mol. Gastroenterol. acquired JAK1 loss-​of-function mutation in mammalian cells using CRISPR interference. Hepatol. 5, 569–590 (2018). driving therapeutic resistance in a patient being Nat. Methods 14, 577–580 (2017). 135. Baratta, M. G. et al. An in-​tumor genetic screen treated with an anti-PD1​ checkpoint inhibitor, 178. Han, K. et al. Synergistic drug combinations for reveals that the BET bromodomain protein, BRD4, is the first example that tumour-​intrinsic genetic cancer identified in a CRISPR screen for pairwise a potential therapeutic target in ovarian carcinoma. mutations play a part in mediating immune genetic interactions. Nat. Biotechnol. 35, 463–474 Proc. Natl Acad. Sci. USA 112, 232–237 (2015). cell killing. (2017). 136. Lord, C. J. & Ashworth, A. The DNA damage response 159. Sade-Feldman,​ M. et al. Resistance to checkpoint 179. Wong, A. S. et al. Multiplexed barcoded CRISPR-​Cas9 and cancer therapy. Nature 481, 287–294 (2012). blockade therapy through inactivation of antigen screening enabled by CombiGEM. Proc. Natl Acad. 137. Muller, F. L. et al. Passenger deletions generate presentation. Nat. Commun. 8, 1136 (2017). Sci. USA 113, 2544–2549 (2016). therapeutic vulnerabilities in cancer. Nature 488, 160. Ozcan, M., Janikovits, J., von Knebel Doeberitz, M. This is the first study to explore the interdependency 337–342 (2012). & Kloor, M. Complex pattern of immune evasion in of two individual genes using a multiplexed This computational analysis of the The Cancer MSI colorectal cancer. Oncoimmunology 7, e1445453 combo-CRISPR​ system. Genome Atlas glioblastoma dataset provides the (2018). 180. Najm, F. J. et al. Orthologous CRISPR-​Cas9 enzymes first examples of the phenomenon of ‘collateral 161. Grasso, C. S. et al. Genetic mechanisms of immune for combinatorial genetic screens. Nat. Biotechnol. 36, damage’, a passenger gene deletion that results in evasion in colorectal cancer. Cancer Discov. 8, 179–189 (2018). a novel dependency, and describes ENO1–ENO2 730–749 (2018). 181. Komor, A. C., Kim, Y. B., Packer, M. S., Zuris, J. A. & in glioblastoma as a putative example. 162. Dupuy, F. et al. LKB1 is a central regulator of Liu, D. R. Programmable editing of a target base in 138. US National Library of Medicine. ClinicalTrials.gov tumor initiation and pro-​growth metabolism in genomic DNA without double-​stranded DNA cleavage. https://clinicaltrials.gov/ct2/show/NCT03435250 (2019). ErbB2-mediated breast cancer. Cancer Metab. 1, 18 Nature 533, 420–424 (2016). 139. Finn, R. S. et al. The cyclin-​dependent kinase 4/6 (2013). 182. Gao, X. et al. Treatment of autosomal dominant inhibitor palbociclib in combination with letrozole 163. Koyama, S. et al. STK11/LKB1 Deficiency promotes hearing loss by in vivo delivery of genome editing versus letrozole alone as first-​line treatment of neutrophil recruitment and proinflammatory cytokine agents. Nature 553, 217–221 (2018). oestrogen receptor-​positive, HER2-negative, advanced production to suppress T cell activity in the lung tumor 183. Nishida, K. et al. Targeted nucleotide editing using breast cancer (PALOMA-1/TRIO-18): a randomised microenvironment. Cancer Res. 76, 999–1008 hybrid prokaryotic and vertebrate adaptive immune phase 2 study. Lancet Oncol. 16, 25–35 (2015). (2016). systems. Science 353, aaf8729 (2016). 140. Finn, R. S. et al. PD 0332991, a selective cyclin D This study is the first to associate loss of 184. Gaudelli, N. M. et al. Programmable base editing of kinase 4/6 inhibitor, preferentially inhibits proliferation function of a known tumour suppressor gene A*T to G*C in genomic DNA without DNA cleavage. of luminal estrogen receptor-​positive human breast (LKB1) with a suppressive immune phenotype Nature 551, 464–471 (2017). cancer cell lines in vitro. Breast Cancer Res. 11, R77 in an engineered mouse tumour model. This 185. Abudayyeh, O. O. et al. C2c2 is a single-​component (2009). finding was subsequently clinically validated in programmable RNA-​guided RNA-targeting​ CRISPR 141. DeMichele, A. et al. CDK 4/6 inhibitor palbociclib the retrospective analysis described in reference effector. Science 353, aaf5573 (2016). (PD0332991) in Rb+ advanced breast cancer: 164. 186. Abudayyeh, O. O. et al. RNA targeting with CRISPR-​ phase II activity, safety, and predictive biomarker 164. Skoulidis, F. et al. STK11/LKB1 mutations and PD-1 Cas13. Nature 550, 280–284 (2017). assessment. Clin. Cancer Res. 21, 995–1001 (2015). inhibitor resistance in KRAS-​mutant lung 187. Cox, D. B. T. et al. RNA editing with CRISPR-​Cas13. 142. Turner, N. C. et al. Palbociclib in hormone-​receptor- adenocarcinoma. Cancer Discov. 8, 822–835 (2018). Science 358, 1019–1027 (2017). positive advanced breast cancer. N. Engl. J. Med. This analysis of a clinical study of patients with 188. Konermann, S. et al. Transcriptome engineering with 373, 209–219 (2015). lung cancer treated with the anti-PD1​ checkpoint RNA-targeting​ type VI-​D CRISPR effectors. Cell 173, 143. Finn, R. S., Aleshin, A. & Slamon, D. J. Targeting inhibitor pembrolizumab is the first to provide 665–676 e614 (2018). the cyclin-dependent​ kinases (CDK) 4/6 in estrogen clinical evidence that LKB1 loss of function has a 189. Yan, W. X. et al. Cas13d is a compact RNA-​targeting receptor-positive​ breast cancers. Breast Cancer Res. significant negative impact on the response to type VI CRISPR effector positively modulated by a 18, 17 (2016). checkpoint inhibitors. WYL-domain-containing​ accessory protein. Mol. Cell 144. Prahallad, A. et al. Unresponsiveness of colon 165. Layer, J. P. et al. Amplification of N-​Myc is associated 70, 327–339.e5 (2018). cancer to BRAF(V600E) inhibition through feedback with a T cell-​poor microenvironment in metastatic 190. Gootenberg, J. S. et al. Multiplexed and portable activation of EGFR. Nature 483, 100–103 (2012). neuroblastoma restraining interferon pathway activity nucleic acid detection platform with Cas13, Cas12a, 145. Kopetz, S. et al. Phase II pilot study of vemurafenib and chemokine expression. Oncoimmunology 6, and Csm6. Science 360, 439–444 (2018). in patients with metastatic BRAF-​mutated colorectal e1320626 (2017). 191. Myhrvold, C. et al. Field-​deployable viral diagnostics cancer. J. Clin. Oncol. 33, 4032–4038 (2015). 166. Kearney, C. J. et al. Tumor immune evasion arises using CRISPR-Cas13.​ Science 360, 444–448 146. Hyman, D. M. et al. Vemurafenib in multiple through loss of TNF sensitivity. Sci. Immunol. 3, (2018). nonmelanoma cancers with BRAF V600 mutations. eaar3451 (2018). 192. Dixit, A. et al. Perturb-​Seq: dissecting molecular N. Engl. J. Med. 373, 726–736 (2015). 167. Peng, W. et al. Loss of PTEN promotes resistance to circuits with scalable single-​cell RNA profiling of 147. Corcoran, R. B. et al. Combined BRAF, EGFR, and T cell-mediated​ immunotherapy. Cancer Discov. 6, pooled genetic screens. Cell 167, 1853–1866.e17 mek inhibition in patients with BRAF(V600E)-mutant 202–216 (2016). (2016). colorectal cancer. Cancer Discov. 8, 428–443 (2018). 168. Hassel, J. C. et al. Combined immune checkpoint This article describes Perturb-Seq,​ a method of 148. Prahallad, A. et al. PTPN11 Is a central node in blockade (anti-PD-1/anti-​ CTLA-4):​ evaluation and using single-cell​ transcriptional output as the intrinsic and acquired resistance to targeted cancer management of adverse drug reactions. Cancer Treat readout for CRISPR-​based functional genomic drugs. Cell Rep. 12, 1978–1985 (2015). Rev. 57, 36–49 (2017). screens.

Nature Reviews | Drug Discovery Reviews

193. Adamson, B. et al. A multiplexed single-​cell CRISPR Acknowledgements grant support from Sun Pharma and AstraZeneca. Patents on screening platform enables systematic dissection of the The authors thank C.P. Johnson and K. Briggs for helpful dis- the use of PARP inhibitors held jointly with AstraZeneca that A.A. unfolded protein response. Cell 167, 1867–1882.e21 cussions and preparation of the manuscript and F. Li and has benefited from financially (and may do so in the future) (2016). T. Teng for curation and analysis of the Project Achilles data- through the ICR Rewards to Inventors Scheme include 194. Jaitin, D. A. et al. Dissecting immune circuits by linking set. All are full-time​ employees of and shareholders in Tango WO2014013231 (A1) — 2014-01-23, US2012135983 (A1) CRISPR-​pooled screens with single-​cell RNA-Seq.​ Cell Therapeutics. — 2012-05-31, US2012010204 (A1) — 2012-01-12, 167, 1883–1896.e15 (2016). US2006142231 (A1) — 2006-06-29, WO2008020180 (A2) 195. Xie, S., Duan, J., Li, B., Zhou, P. & Hon, G. C. Multiplexed Author contributions — 2008-02-21. engineering and analysis of combinatorial enhancer B.W. and A.H. researched data for article and wrote the arti- activity in single cells. Mol. Cell 66, 285–299.e5 cle. A.A. contributed to writing the article. All authors contrib­ Publisher’s note (2017). uted substantially to discussion of the content and reviewed Springer Nature remains neutral with regard to jurisdictional 196. Datlinger, P. et al. Pooled CRISPR screening with and edited the manuscript before submission: claims in published maps and institutional affiliations. single-cell​ transcriptome readout. Nat. Methods 14, 297–301 (2017). Competing interests Related links 197. Mirza, M. R. et al. Niraparib maintenance therapy in A.H. and B.W. are employees of and shareholders in Tango Cancer Cell Line encyclopedia: https://portals. platinum-​sensitive, recurrent ovarian cancer. N. Engl. Therapeutics. L.A.G. is an employee of and shareholder in Eli Lilly broadinstitute.org/ccle J. Med. 375, 2154–2164 (2016). and Company and is a shareholder in Tango Therapeutics. A.A. Project Achilles: https://depmap.org/portal/achilles/ 198. Li, W. et al. MAGeCK enables robust identification is a shareholder in Tango Therapeutics, a consultant for Project DRive: https://oncologynibr.shinyapps.io/drive/ of essential genes from genome-​scale CRISPR/Cas9 AtlasMDX, Third Rock Ventures, Pfizer, ProLynx and Bluestar Project score: https://score.depmap.sanger.ac.uk/ knockout screens. Genome Biol. 15, 554 (2014). and a Genentech scientific advisory board member and receives

www.nature.com/nrd